Uncle Bob: My current strategy is to not read any code written by my agents

ivm1 pts0 comments

Uncle Bob Martin (@unclebobmartin): "I’m significantly older than you. I started coding in the late 60s. My current strategy is to not read any of the code written by my agents. That’s the only way I can take advantage of their productivity. What I do instead is to surround the agents with extreme constraints. Unit tests, gherkin tests, QA procedures, quality metrics, mutation testing, test coverage, and a plethora of others. In the end, I have very high confidence in the code they produce because they’ve had to run the gauntlet of all of my constraints and tests." | XCancel

Ori Pomerantz@ori_pomerantz

Jul 22

I am trying to use Claude to help me write something, but I just don't feel comfortable letting it edit my files. Does anybody else feel the same? If I am responsible for code, I NEED to understand it, psychologically if for no other reason.

Started programming in 1983. Old?

205

323

83,482

Uncle Bob Martin

@unclebobmartin

17h

Replying to @ori_pomerantz

I’m significantly older than you. I started coding in the late 60s. My current strategy is to not read any of the code written by my agents. That’s the only way I can take advantage of their productivity. What I do instead is to surround the agents with extreme constraints. Unit tests, gherkin tests, QA procedures, quality metrics, mutation testing, test coverage, and a plethora of others. In the end, I have very high confidence in the code they produce because they’ve had to run the gauntlet of all of my constraints and tests.

Jul 23, 2026 · 11:44 AM UTC

296

849

9,363

1,689,756

Sort replies:<br>Relevant<br>Recent<br>Liked

Wahyu

@WahyuCodes

9h

Replying to @unclebobmartin @ori_pomerantz

i am trying to do the same exact things, I am curious, what deterministic tools you can recommend for quality metrics? because that is such a non deterministic thing to measure.

35

25,031

Uncle Bob Martin

@unclebobmartin

8h

GitHub.com/unclebob/crap4jav….

GitHub - unclebob/crap4java: Crap analyzer for my java projects.

Crap analyzer for my java projects. . Contribute to unclebob/crap4java development by creating an account on GitHub.

github.com

221

21,306

monkeypatch@m0nkeypatch

17h

Replying to @unclebobmartin @ori_pomerantz

While you're ensuring product quality through testing, what about the quality of the code itself? Lately I have been wondering, that does it even matter anymore, now that changing code is so cheap and easy? Or am I missing something?

24

25,925

Uncle Bob Martin

@unclebobmartin

15h

I think it matters still, and I think it matters a lot. Messy code slows my agents down. I've seen them wrangle with their own messes without resolution. I finally had to step in and untangle their own mess. So I don't let them create those tangles. I constrain the hell out of function sizes, cyclomatic complexity, and test coverage. That seems to keep them moving smoothly.

228

22,969

more replies

Prahlad Yeri@prahladyeri

16h

Replying to @unclebobmartin @ori_pomerantz

By doing so, you've trapped yourself into the Chinese room argument. If the LLM generates all the code and you're merely testing it with constraints, then are you really the engineer (Chinese translator) here?

10

40

24,618

Uncle Bob Martin

@unclebobmartin

16h

I am the engineer because I am accountable.

18

35

697

25,814

Daniel Redding@theRedding

9h

Replying to @unclebobmartin @ori_pomerantz

What’s your favorite type of thing to have them build?

I don’t know any coding but so far I’ve got a personal context social media app, a semantic suggestion writing and notes app, and an IR remote made from an ESP32 that is controlled by Siri.

5,595

Uncle Bob Martin

@unclebobmartin

8h

Mostly games and tools.

13

4,333

Marko@MarkoVelich

9h

Replying to @unclebobmartin @ori_pomerantz

Works fine until they start changing tests :)

4,492

Uncle Bob Martin

@unclebobmartin

8h

That’s why we overload the tests with mutation testing, QA procedures, gherkin tests, and so on. It’s not just one test they have to change. it’s a lot of tests. And that makes it harder to change the tests.

21

4,104

Repojournal

@repojournal

8h

Replying to @unclebobmartin @ori_pomerantz

If we cannot trust the outcome of what AI makes and AI has proven over and over again that it will not listen to us at all times what gives us the confidence that it will stay within the guardrails of our constraints just because we asked it to?

If AI simply makes the judgement call to change the test to what it wants or to have false positives how do we know?

Hidden in between the test checks, haven’t we seen bugs where test pass and QA passes too?

In our writing of code how many times have we seen cases where our assumptions simply turned out to be wrong and we need to take a different direction entirely. How do we handle these sorts of cases with AI?

Are comfortable walking in a future where we cannot say what a software does? We cannot say a software does something simply because the test we...

unclebobmartin code tests uncle test ori_pomerantz

Related Articles